Page 1 of 1

Negative deviance degrees of freedom

PostPosted: Tue Feb 28, 2017 8:48 am
by GDistiller
What can cause the reported number for the "Deviance degrees of freedom" to be negative?

I have a sparse dataset with 5 occasions and 10 different types of capture histories and after fitting a fully time-dependent CJS model the output reports the deviance dof as -2 and the c-hat estimate as 1.00. One thought I had was that the model is saturated but as I understand things there are effectively 10 data points and only 7 parameters that are estimated. I have played with the data by adding in new rows to the INP file (by just making up some new capture histories) and seen that if I add three new rows to the data I then get a reported deviance df of 1 and an estimated c-hat value.

All I can think is that there are very few recaptures in the data and it is not only the number of capture histories but also the structure of those capture histories that affects the deviance df. Can anyone shed any further light on how the degrees of freedom are calculated?



Re: Negative deviance degrees of freedom

PostPosted: Fri Nov 03, 2017 11:18 am
by aswea
I'm interested to know the answer to this as well.